The Global Open Source Severity of Illness Score (GOSSIS)

Area of Science:

• Critical Care Medicine
• Health Informatics
• Biostatistics

Background:

• Existing severity of illness scores often lack generalizability across diverse healthcare systems and countries.
• There is a need for a modern, open-source tool for predicting inhospital mortality in intensive care units (ICUs).

Purpose of the Study:

• To develop and validate the Global Open Source Severity of Illness Score (GOSSIS)-1 for critical care patients.
• To demonstrate the feasibility of a generalized score across international healthcare settings.

Main Methods:

• A merged dataset from multicenter cohorts (ANZICS-APD and eICU-CRD) was used, split into 70% training and 30% testing sets.
• GOSSIS-1 incorporates data from the first 24 hours of ICU stay, including vital signs, lab results, diagnoses, and comorbidities.
• Missing data were imputed to ensure data completeness for model development and evaluation.

Main Results:

• GOSSIS-1 achieved high discrimination (AUROC, 0.918) and excellent calibration (SMR, 0.986; Brier score, 0.050) on the test set.
• The score performed consistently well within both the ANZICS-APD (AUROC, 0.925) and eICU-CRD (AUROC, 0.904) datasets.
• GOSSIS-1 outperformed existing scores like APACHE-IIIj and APACHE-IVa in discrimination and calibration.

Conclusions:

• GOSSIS-1 is a validated, free, and open-source algorithm for predicting inhospital mortality in critical care.
• The score demonstrates excellent generalizability and performance across different countries and healthcare systems.
• This tool offers a modern, accessible solution for severity of illness assessment in ICUs worldwide.